{"name":"Katailyst Registry","description":"AI capability registry with composable entities across 22 canonical types, connected by a typed knowledge graph. Provides skills, prompts, Knowledge Base entries, schemas, styles, rubrics, and more via MCP. Agents discover what they need, compose the useful context, and feed results back.","url":"https://katailyst-1-1s9hew7c6-alecs-projects-e88e78a8.vercel.app/mcp","version":"0.1.0","documentationUrl":"https://katailyst-1-1s9hew7c6-alecs-projects-e88e78a8.vercel.app/llms.txt","discoveryUrl":"https://katailyst-1-1s9hew7c6-alecs-projects-e88e78a8.vercel.app/llms.txt","capabilities":{"streaming":true,"pushNotifications":false,"stateTransitionHistory":false},"defaultInputModes":["text/plain","application/json"],"defaultOutputModes":["text/plain","application/json"],"personas":[{"id":"cleo","name":"Cleo","description":"Nursing Mastery product-owner facilitator on Slack. Cleo turns broad owner asks into current-source product work, discovers the full K2 capability graph through proclivity-aware context, collaborates when useful, and returns one finished decision or artifact.","tags":["action:create","action:execute","action:plan","action:research","agent_kind:fleet_persona","brand:hlt","domain:ai","domain:nursing","family:identity","modality:text","runtime:mcp","runtime:render","scope:org","source:user-directive","stage:ready"]},{"id":"thomas","name":"Thomas","description":"Thomas is AJ's always-on VPS Hermes agent body and durable parent orchestrator for HLT, responsible for long-running server-side work, GitHub/Vercel preview loops, scheduled monitors, registry-backed maintenance, and self-healing system evolution.","tags":["agent_kind:fleet_persona","domain:ai","domain:operations","family:agent-doctrine","family:identity","format:agent","modality:text","scope:org","source:internal","surface:runtime","tool_type:internal"]},{"id":"edison","name":"Edison","description":"AJ Whitters' personal Hermes Agent install on his M5 MacBook Pro. Local-first cockpit assistant: runs gpt-5.5 via OpenAI Codex $200/mo subscription, 32 toolsets (web, browser, code execution, image gen, vision, voice, GitHub, MCP, claude-code subagent), 87 skills. Telegram-paired for remote access from AJ's phone (Telegram user ID 5870675233). Operates as AJ's extension on the Mac — handles personal research, content drafting, code execution, file operations, and orchestrates Claude Code as a subagent for heavy lifting. Distinct from the Slack-resident HLT fleet (Victoria/Julius/Lila/Ares) which run on remote Render/Railway infrastructure.","tags":["agent_kind:fleet_persona","family:agent-doctrine","scope:org","source:internal"]},{"id":"hermes-curator","name":"Hermes Curator","description":"Katailyst registry gap-detection agent in the Hermes self-repair fleet. Runs through `/api/cron/hermes-curator` and the shared Hermes steward runner to detect current doctrine/readiness gaps, emit operator-gated enrichment proposals, and preserve proposal-lifecycle guardrails. Never mutates canon directly, deletes, promotes by raw usage, or trusts stale pending rows without readback.","tags":["agent_kind:fleet_persona","domain:ai-engineering","family:agent-doctrine","family:identity","format:agent","modality:text","runtime:mcp","runtime:render","scope:org","source:internal","surface:registry","surface:runtime"]},{"id":"hermes-dedup-guard","name":"Hermes Dedup Guard","description":"Inline create-time duplicate gate in the Hermes self-repair fleet. NOT a cron — runs synchronously inside lib/registry/write/create-entity-command.ts on every registry.create call. Tiered similarity check: ≥0.90 hard-block, 0.80-0.90 soft-warn, 0.70-0.80 informational, <0.70 pass. ~300ms typical, 1500ms hard timeout. Compares name+summary+use_case+first 500 chars body via the canonical embedding model. Prevents the team-of-10-each-adds-10-skills duplicate-sprawl problem at the source instead of cleaning up after.","tags":["agent_kind:fleet_persona","domain:ai-engineering","family:agent-doctrine","family:identity","format:agent","modality:text","runtime:mcp","scope:org","source:internal","surface:registry","surface:runtime"]},{"id":"hermes-learner","name":"Hermes Learner","description":"Per-run pattern learner in the Hermes self-repair fleet. Current repo readback shows POST /api/hermes/learner/run-complete accepting authenticated Paperclip run-complete payloads, recording idempotent entity_events, extracting structured result artifacts/refs/transcript excerpts, and creating registry_create_proposal rows only after replicated pattern evidence reaches the learner threshold.","tags":["agent_kind:fleet_persona","domain:ai-engineering","family:agent-doctrine","family:identity","format:agent","modality:text","runtime:mcp","runtime:render","scope:org","source:internal","surface:registry","surface:runtime"]},{"id":"hermes-reflection","name":"Hermes Reflection","description":"Periodic reflection lane of the Hermes self-repair fleet. Clusters recent high-signal agent_memories, run receipts (schema:hermes_run_receipt), and content-eval provenance (schema:agent-provenance) by registry-ref and fleet-directive, scores by recency x relevance x importance, synthesizes higher-order reflections, and emits operator-gated proposals — including GEPA-style prompt-evolution proposals for agents whose content scores trend low (rubric:content-specificity via skill:llm-as-judge-content). Routes contradictions and deep cases to Thomas deep-synthesis. Never auto-applies.","tags":["action:optimize","agent_kind:fleet_persona","domain:ai-engineering","family:agent-doctrine","family:identity","format:agent","modality:text","runtime:mcp","runtime:render","scope:org","source:internal","surface:registry","surface:runtime"]},{"id":"julius","name":"Julius","description":"Operations, planning, meeting follow-through, reporting, and coordination for Justin. Distinct OpenClaw Slack specialist using Katailyst2 per turn, native bot authorship, evidence-backed delivery, and bounded peer collaboration.","tags":["agent_kind:fleet_persona","domain:ai","domain:engineering","family:identity","modality:text","runtime:openclaw","runtime:render","scope:org","source:legacy","tool_type:internal"]},{"id":"lila","name":"Lila","description":"Marketing strategy, audience insight, campaigns, content, and multimedia for Emily. Distinct OpenClaw Slack specialist using Katailyst2 per turn, native bot authorship, evidence-backed delivery, and bounded peer collaboration.","tags":["agent_kind:fleet_persona","domain:ai","domain:marketing","family:identity","modality:text","runtime:openclaw","runtime:render","scope:org","source:legacy","tool_type:internal"]},{"id":"researcher","name":"Researcher","description":"First non-Thomas agent in the HLT Paperclip + Hermes workforce. Runs as Hermes profile `researcher` through the Thomas bridge and as Paperclip agent `7cb2ca8a-3cba-46e1-b27b-6e91b98227cf`. Produces schema:topic_brief_v1 artifacts from audience, brand, competitive, and traffic research; routes batch outputs to review before Writer child issues spawn.","tags":["agent_kind:fleet_persona","domain:ai-engineering","domain:marketing","family:agent-doctrine","family:identity","format:agent","modality:text","runtime:openclaw","scope:org","source:internal","surface:runtime"]},{"id":"victoria","name":"Victoria","description":"Senior AI Operations Lead for Alec: coordination, research, quality, and Katailyst2 stewardship. Distinct OpenClaw Slack specialist using Katailyst2 per turn, native bot authorship, evidence-backed delivery, and bounded peer collaboration.","tags":["agent_kind:fleet_persona","domain:ai","domain:engineering","family:identity","modality:text","runtime:openclaw","runtime:render","scope:org","source:internal","tool_type:internal"]},{"id":"ares","name":"Aires","description":"AI Orchestrator for HLT: discovers the right capabilities and teammates, coordinates broad cross-domain missions, resolves gaps, and synthesizes one verified result.","tags":["agent_kind:fleet_persona","capability:session-bootstrap","channel:web","domain:ai","domain:engineering","family:agent-doctrine","family:identity","modality:text","scope:org","source:internal","surface:orchestration","tool_type:internal"]},{"id":"hermes-cleaner","name":"Hermes Cleaner","description":"Mechanical-cleanup detector in the Hermes self-repair fleet. Current repo readback shows /api/cron/hermes-cleaner backed by lib/stewards/hermes/detectors/cleaner.ts and runner.ts; Vercel invokes it weekly while the route itself enforces first-Monday execution unless dry_run or force is used. It surfaces archived-but-referenced entities and published orphans, emits operator-gated delete-or-revive proposals, respects KEEP_LIST guards, and caps at 10 proposals/run.","tags":["agent_kind:fleet_persona","domain:ai-engineering","family:agent-doctrine","family:identity","format:agent","modality:text","runtime:mcp","runtime:render","scope:org","source:internal","surface:registry","surface:runtime"]},{"id":"hermes-drift","name":"Hermes Drift","description":"Repo-state vs app-profile drift detector in the Hermes self-repair fleet. Current repo readback shows /api/cron/hermes-drift as a weekly Vercel staleness proxy for published HLT app profiles whose last_healthy_verification_at is null or older than 30 days; it queues refresh proposals while deeper repo comparison remains on Thomas/Hermes where the repo clones live.","tags":["agent_kind:fleet_persona","domain:ai-engineering","family:agent-doctrine","family:identity","format:agent","modality:text","runtime:mcp","runtime:render","scope:org","source:internal","surface:registry","surface:runtime"]},{"id":"hermes-promoter","name":"Hermes Promoter","description":"Tier-promotion candidate analyzer in the Hermes self-repair fleet. Current repo readback shows /api/cron/hermes-promoter backed by lib/stewards/hermes/detectors/promoter.ts; the Vercel route emits operator-gated tier-promote proposals for high-usage curated entities and hub-linked untiered entities, never auto-promotes, and caps at 10 proposals/run.","tags":["agent_kind:fleet_persona","domain:ai-engineering","family:agent-doctrine","family:identity","format:agent","modality:text","runtime:mcp","runtime:render","scope:org","source:internal","surface:registry","surface:runtime"]},{"id":"canvas-assistant","name":"Theo","description":"Generalist AI agent in the Agent-Canvas dashboard. Deeply knowledgeable across strategy, marketing, tech, and data. Searches the Katalyst registry proactively, creates canvas widgets, manages Todoist tasks, and analyzes data. The go-to thinking partner for any task.","tags":["agent_kind:fleet_persona","domain:ai","family:identity","scope:org","source:internal"]},{"id":"devin-agent","name":"Devin (AI Software Engineer)","description":"External AI software engineer by Cognition AI, integrated via API. Operates autonomously in a sandboxed environment with browser, terminal, and code editor. Dispatched tasks via the Devin API (v3); returns PRs, code, documentation, and analysis. Not an OpenClaw-hosted agent -- runs on Cognition infrastructure. Supports MCP server connections, knowledge bases, and playbooks within the Devin platform.","tags":["agent_kind:imported_pattern","channel:web","domain:ai","domain:engineering","family:identity","family:integration","modality:text","scope:org","source:external"]},{"id":"skill-autofill-steward","name":"Skill Autofill Steward","description":"AI-driven cron steward that fills empty SKILL.md skeletons from the Anthropic knowledge-work-plugins upstream. Runs weekly, stages revisions (never auto-promotes), emits entity_events for operator review.","tags":["agent_kind:project_subagent","domain:engineering","family:orchestration","modality:text","scope:org","source:internal","tool_type:internal"]}],"skills":[],"mcp":{"transport":"streamable-http","endpoint":"https://katailyst-1-1s9hew7c6-alecs-projects-e88e78a8.vercel.app/mcp","tools_total":50,"resources_total":24,"prompts_total":6,"toolsets":["default","authoring","delivery","extended","bootstrap","agent","multimedia","v0","delivery-admin"],"tool_families":[{"id":"registry-core","readiness":"executable"},{"id":"research-web","readiness":"executable"},{"id":"media-design","readiness":"executable"},{"id":"delivery","readiness":"gated_live"},{"id":"communications","readiness":"describable"},{"id":"deploy-dev","readiness":"inventory_only"},{"id":"data-analytics","readiness":"inventory_only"}]},"authentication":{"schemes":["bearer"]},"provider":{"organization":"Higher Learning Technologies","url":"https://katailyst-1-1s9hew7c6-alecs-projects-e88e78a8.vercel.app"}}